Llama-3.1-8B-Instruct_dpo_sg_values_p025_OA_gold

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the dpo_sg_values_p025_OA_gold dataset. It achieves the following results on the evaluation set:

• Loss: 0.1369
• Rewards/chosen: -0.3243
• Rewards/rejected: -3.7445
• Rewards/accuracies: 0.9400
• Rewards/margins: 3.4202
• Logps/chosen: -5.3343
• Logps/rejected: -44.0676
• Logits/chosen: -0.8235
• Logits/rejected: -0.8529

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/chosen	Logps/rejected	Logits/chosen	Logits/rejected
0.677	0.1495	250	0.6611	-0.0030	-0.0688	0.8780	0.0658	-2.1210	-7.3104	-0.6673	-0.6792
0.4141	0.2990	500	0.3328	-0.1370	-1.2934	0.8960	1.1564	-3.4617	-19.5564	-0.7405	-0.7604
0.1869	0.4486	750	0.1943	-0.3266	-2.7983	0.9280	2.4718	-5.3572	-34.6058	-0.8272	-0.8525
0.1234	0.5981	1000	0.1579	-0.3430	-3.2984	0.9380	2.9554	-5.5213	-39.6065	-0.8336	-0.8610
0.122	0.7476	1250	0.1439	-0.3187	-3.5609	0.9360	3.2422	-5.2784	-42.2310	-0.8265	-0.8553
0.0821	0.8971	1500	0.1398	-0.3263	-3.7101	0.9340	3.3838	-5.3544	-43.7234	-0.8241	-0.8535

Framework versions

• PEFT 0.15.2
• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 2.21.0
• Tokenizers 0.21.1